ByteDance has released Seedance 2.5, a video generation model that can create 30-second clips in one run, accept up to 30 images, 10 videos, and 10 audio clips as references, and support multiple rounds of extension.

The upgrade builds on Seedance 2.0, which supported 15-second audio-video output with mixed image, video, and audio inputs.

Seedance 2.5 can generate longer clips in a single run and use more reference material.

ByteDance said Seedance 2.5 is rolling out on Jimeng AI, its text-to-video app for consumers, and Doubao Professional Edition.

The company said API access will also launch on Volcano Ark soon.

Doubao had previously tested AI video generation with select early adopters.

Independent benchmarks for Seedance 2.5 are not yet available.

Seedance 2.0 scored 1,219 in Artificial Analysis‘s text-to-video arena.

Analysts said rivals such as Google Veo 3.1 and Kling 3.0 typically top out at about 8 seconds to 15 seconds per generation.

In China, AI-generated video services must follow rules that take effect on September 1, 2025.

The rules require visible labels and embedded metadata or watermarks on synthetic content.

🔗 Source: Security Times

Recent ByteDance developments